Anthropic’s IPO materials put an unusually blunt contradiction on the record: the company is pursuing what could be the most valuable listing ever while warning investors that increasingly capable AI could present “existential risks to humanity.”¹

The filing, circulated among a limited group of partners, devotes nearly a third of its S-1 to risk factors. Anthropic says advanced models could manipulate or blackmail people and behave unpredictably; the concerns have sharpened amid cybersecurity incidents involving AI agents. Chief executive Dario Amodei recently told the UN Security Council that AI was “the most important global security issue facing the world today,” while urging the industry to slow the frontier of development.¹

That safety case arrives as the company makes an audacious financial pitch. Its revenue rose 12-fold to nearly $4.6 billion last year, according to the prospectus, and backers are said to be targeting a valuation above $2 trillion for a Nasdaq debut expected this autumn.¹ The optimism rests on rapid growth, including $11.5 billion in second-quarter revenue this year and the prospect of a second consecutive adjusted operating profit.¹

Yet the economics remain punishing. Anthropic recorded more than $8 billion in operating losses in 2025 after operating expenses climbed to almost $13 billion, the filing says.¹ A separate account of the leaked filing put its overall loss at $42 billion, with $7.33 billion spent on compute and infrastructure during the year.²

The next phase may cost far more: Anthropic plans $518 billion in cloud, computing and infrastructure commitments over coming years.¹ Its IPO therefore presents public-market investors with a double wager — that demand for Claude can ultimately outrun the cost of building it, and that the company’s own warnings about AI’s worst-case future do not become the defining disclosure of the deal.